Image Retrieval on Stanford Online Products (test)
88.3Recall@1STIR-Symmetric
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| STIR-SymmetricEmbedding Size=384, Backbone=ViT-Triplet, n=5, Approach=Reranking approaches2023.04 | 88.3 | 95.3 | 98.1 | — | — | — | |
| CHEST-ViTArch (dim)=ViT-B-E (1024), Space=Euclidean2025.10 | 88.2 | 95.8 | 98.1 | — | — | 67.3 | |
| STIREmbedding Size=384, Backbone=ViT-Triplet, n=5, Approach=Reranking approaches2023.04 | 88.1 | 95.3 | 98.1 | — | — | — | |
| RS@KArch (dim)=ViT-B (512)2025.10 | 88 | 96.1 | 98.6 | — | — | — | |
| CHEST-ViTArch (dim)=ViT-B-H (512), Space=Hyperbolic2025.10 | 88 | 95.6 | 97.9 | — | — | 67.1 | |
| LaFGArch=ViT2025.12 | 87.1 | 95.8 | 98.5 | 99.5 | — | — | |
| VPTSP-GArch (dim)=ViT-B (512)2025.10 | 86.8 | 95 | 98 | — | — | — | |
| ViT-TripletEmbedding Size=384, Approach=Embedding-based retrieval2023.04 | 86.5 | 95.2 | 98.1 | — | — | — | |
| CHEST-ViTArch (dim)=ViT-S-H (384), Space=Hyperbolic2025.10 | 86.5 | 94.7 | 97.6 | — | — | 64.2 | |
| SEEArch=ViT2025.12 | 86.3 | 95 | 98.2 | — | — | — | |
| CHEST-ViTArch (dim)=ViT-S-E (384), Space=Euclidean2025.10 | 86.3 | 94.7 | 97.5 | — | — | 63.7 | |
| HIERArch=ViT2025.12 | 86.1 | 95 | 98 | — | — | — | |
| DDMLArch=ViT2025.12 | 86.1 | 95.1 | 98.2 | 99.5 | — | — | |
| HIERArch (dim)=ViT-S (384)2025.10 | 86.1 | 95 | 98 | — | — | — | |
| ROADMAPEmbedding Size=384, Backbone=DeiT-s, Approach=Embedding-based retrieval2023.04 | 86 | 94.4 | 97.6 | — | — | — | |
| HypVitEmbedding Size=384, Approach=Embedding-based retrieval2023.04 | 85.9 | 94.9 | 98.1 | — | — | — | |
| HypViTArch=ViT2025.12 | 85.9 | 94.9 | 98.1 | 99.5 | — | — | |
| Hyp-ViTArch (dim)=ViT-S (384)2025.10 | 85.9 | 94.9 | 98.1 | — | — | — | |
| Surrogate recallEmbedding Size=384, Backbone=ViTs16, Approach=Embedding-based retrieval2023.04 | 85.6 | 94.8 | 98 | — | — | — | |
| HypVitEmbedding Size=128, Approach=Embedding-based retrieval2023.04 | 85.5 | 94.9 | 98.1 | — | — | — | |
| Hyp-DINOEmbedding Size=384, Approach=Embedding-based retrieval2023.04 | 85.1 | 94.4 | 97.8 | — | — | — | |
| DPHMArch=ViT2025.12 | 84.8 | 94.5 | 98.1 | 99.4 | — | — | |
| Hyp-DINOEmbedding Size=128, Approach=Embedding-based retrieval2023.04 | 84.6 | 94.1 | 97.7 | — | — | — | |
| CO + RRT (finetuned)Descriptor Dimension=128, Mode=Reranking (finetuned backbone)2021.03 | 84.5 | 93.2 | 96.6 | 99 | — | — | |
| Reranking TransformersProtocol=finetuned, Approach=Reranking approaches2023.04 | 84.5 | 93.2 | 96.6 | — | — | — | |
| VPTSP-GArch=ViT2025.12 | 84.4 | 93.6 | 97.3 | — | — | — | |
| VPTSP-GArch (dim)=ViT-S (384)2025.10 | 84.4 | 93.6 | 97.3 | — | — | — | |
| CGDBackbone=SE-ResNet-50, Dim=1536, Input Size=256x256, Configuration=SG2019.03 | 84.2 | 93.9 | 97.4 | 99.2 | — | — | |
| IRTREmbedding Size=384, Backbone=DeiT-s, Approach=Embedding-based retrieval2023.04 | 84.2 | 93.7 | 97.3 | — | — | — | |
| DFML-PAArch=ViT2025.12 | 84.2 | 93.8 | — | — | — | — | |
| CGDBackbone=ResNet-50, Dim=1536, Input Size=256x256, Configuration=SG2019.03 | 83.9 | 93.8 | 97.5 | 99.2 | — | — | |
| LM-MetricArch=ViT2025.12 | 83.1 | — | — | — | — | — | |
| BFE+Backbone=ResNet-50, Dim=1536, Input Size=256x2562019.03 | 83 | 93.3 | 97.3 | 99.2 | — | — | |
| CO + RRT (frozen)Descriptor Dimension=128, Mode=Reranking (frozen backbone)2021.03 | 81.8 | 92.4 | 96.6 | 99 | — | — | |
| Reranking TransformersProtocol=frozen, Approach=Reranking approaches2023.04 | 81.8 | 92.4 | 96.6 | — | — | — | |
| HISTArch=R502025.12 | 81.4 | 92 | 96.7 | — | — | — | |
| PNCA++Arch=R502025.12 | 81.4 | 92.4 | 96.9 | 99 | — | — | |
| Cross-Entropyd=cos, Architecture=ResNet-502020.03 | 81.1 | 91.7 | 96.3 | 98.8 | — | — | |
| PNP-DqBatch Size=3842021.02 | 81.1 | 92.2 | 96.8 | 99 | — | — | |
| CEDescriptor Dimension=2048, Type=Global-only retrieval2021.03 | 81.1 | 91.7 | 96.3 | 98.8 | — | — | |
| CGDBackbone=ResNet-50, Dim=128, Input Size=224x224, Configuration=SG2019.03 | 81 | 92.2 | 96.8 | 99.1 | — | — | |
| Cross-Entropyd=l2, Architecture=ResNet-502020.03 | 80.8 | 91.2 | 95.7 | 98.1 | — | — | |
| Pnca++2021.02 | 80.7 | 92 | 96.7 | 98.9 | — | — | |
| CODescriptor Dimension=128, Type=Global-only retrieval2021.03 | 80.7 | 91.9 | 96.6 | 99 | — | — | |
| Cont. w/MBackbone=ResNet502020.07 | 80.6 | 91.6 | 96.2 | 98.7 | — | — | |
| Cont. w/M2021.02 | 80.6 | 91.6 | 96.2 | 98.7 | — | — | |
| XBMDescriptor Dimension=128, Type=Global-only retrieval2021.03 | 80.6 | 91.6 | 96.2 | 98.7 | — | — | |
| CGDBackbone=BN-Inception, Dim=512, Input Size=224x224, Configuration=SG2019.03 | 80.5 | 92.1 | 96.7 | 98.9 | — | — | |
| MetricOpt + Recall lossBackbone=ResNet-50, Embedding size=512, Batch size=128, Loss Function=MetricOpt with Recall loss surrogate, FiLM conditioning parameters=128, Finetuning steps=2002021.04 | 80.4 | 91.8 | 96.5 | 99 | — | — | |
| NIRArch=R502025.12 | 80.4 | 91.4 | — | — | — | — | |
| Proxy-AnchorEmbedding size=512, larger input image=true2020.03 | 80.3 | 91.4 | 96.4 | 98.7 | — | — | |
| Contra+HORDEEmbedding size=512, larger input image=true2020.03 | 80.1 | 91.3 | 96.2 | 98.7 | — | — | |
| HORDEBackbone=I3, Embedding Dimension=5122020.04 | 80.1 | 91.3 | 96.2 | 98.7 | — | — | |
| Smooth-APBackbone=ResNet50, mini-batch size=3842020.07 | 80.1 | 91.5 | 96.6 | 99 | — | — | |
| Smooth-AP2021.02 | 80.1 | 91.5 | 96.6 | 99 | — | — | |
| PNP-DqBatch Size=1122021.02 | 80.1 | 91.5 | 96.7 | 99 | — | — | |
| HSEArch=R502025.12 | 80 | 91.4 | 96.3 | — | — | — | |
| DRMLSetting=512R2022.03 | 79.9 | 90.7 | 96.1 | — | — | — | |
| CBMLArch=R502025.12 | 79.9 | 91.5 | 96.5 | 98.9 | — | — | |
| DCMLSetting=512R2022.03 | 79.8 | 90.8 | 95.8 | — | — | — | |
| DCML-MDW2021.02 | 79.8 | 90.8 | 95.8 | — | — | — | |
| MetricOpt + FastAPBackbone=ResNet-50, Embedding size=512, Batch size=128, Loss Function=MetricOpt with FastAP surrogate, FiLM conditioning parameters=128, Finetuning steps=2002021.04 | 79.6 | 91.2 | 96.3 | 98.9 | — | — | |
| ProxyAnchor-AVSLSetting=512R2022.03 | 79.6 | 91.4 | 96.4 | — | — | — | |
| NormSoftmaxBackbone=ResNet-50, Embedding size=20482018.11 | 79.5 | 91.5 | 96.7 | — | — | — | |
| NormSoftMaxBackbone=R50, Embedding Dimension=20482020.04 | 79.5 | 91.5 | 96.7 | — | — | — | |
| NormSoftmaxd=cos, Architecture=ResNet-502020.03 | 79.5 | 91.5 | 96.7 | — | — | — | |
| DIMLArch=ViT2025.12 | 79.5 | — | — | — | — | — | |
| DIMLSetting=512R2022.03 | 79.3 | — | — | — | — | — | |
| Smooth-APBackbone=ResNet50, mini-batch size=2242020.07 | 79.2 | 91 | 96.5 | 98.9 | — | — | |
| Proxy-AnchorEmbedding size=5122020.03 | 79.1 | 90.8 | 96.2 | 98.7 | — | — | |
| CGDBackbone=ShuffleNet-v2, Dim=1536, Input Size=224x224, Configuration=SG2019.03 | 78.7 | 90.9 | 96.1 | 98.8 | — | — | |
| Blackbox APBackbone=ResNet502020.07 | 78.6 | 90.5 | 96 | 98.7 | — | — | |
| Recall lossBackbone=ResNet-50, Embedding size=512, Batch size=128, Loss Function=Recall loss surrogate2021.04 | 78.6 | 90.5 | 96 | 98.7 | — | — | |
| Blackbox AP2021.02 | 78.6 | 90.5 | 96 | 98.7 | — | — | |
| ProxyAnchorSetting=512R2022.03 | 78.4 | 90.5 | 96.2 | — | — | — | |
| SoftTripleEmbedding size=5122020.03 | 78.3 | 90.3 | 95.9 | — | — | — | |
| SoftTripleEmbedding Dimension=large2019.09 | 78.3 | 90.3 | 95.9 | — | 92 | — | |
| CircleLossEmbedding size=5122020.02 | 78.3 | 90.5 | 96.1 | 98.6 | — | — | |
| EPSHNd=cos, Architecture=ResNet-502020.03 | 78.3 | 90.7 | 96.3 | — | — | — | |
| SoftTripleSetting=512BN2022.03 | 78.3 | 90.3 | 95.9 | — | — | — | |
| CircleSetting=512BN2022.03 | 78.3 | 90.5 | 96.1 | — | — | — | |
| SoftTriplet2021.02 | 78.3 | 90.3 | 95.9 | — | — | — | |
| SoftTripleArch (dim)=I (512)2025.10 | 78.3 | 90.3 | 95.9 | — | — | — | |
| NormSoftmaxBackbone=ResNet-50, Embedding size=5122018.11 | 78.2 | 90.6 | 96.2 | — | — | — | |
| NormSoftmaxBackbone=ResNet-50, Embedding size=2048bits2018.11 | 78.2 | 90.9 | 96.4 | — | — | — | |
| MSEmbedding size=5122019.04 | 78.2 | 90.5 | 96 | 98.7 | — | — | |
| MSEmbedding size=5122020.03 | 78.2 | 90.5 | 96 | 98.7 | — | — | |
| Multi-SimiEmbedding size=5122020.02 | 78.2 | 90.5 | 96 | 98.7 | — | — | |
| MSBackbone=I3, Embedding Dimension=5122020.04 | 78.2 | 90.5 | 96 | 98.7 | — | — | |
| NormSoftMaxBackbone=R50, Embedding Dimension=5122020.04 | 78.2 | 90.6 | 96.2 | — | — | — | |
| Multi-Similarityd=cos, Architecture=BN-Inception2020.03 | 78.2 | 90.5 | 96 | 98.7 | — | — | |
| MSSetting=512BN2022.03 | 78.2 | 90.5 | 96 | — | — | — | |
| MPAArch (dim)=I (512)2025.10 | 78.1 | 90.1 | 95.6 | — | — | 50.2 | |
| SoftMax_normEmbedding Dimension=large2019.09 | 78 | 90.2 | 96 | — | 91.9 | — | |
| LMCL*Backbone=ResNet-50, Embedding size=20482018.11 | 77.4 | 89.7 | 95.3 | — | — | — | |
| MICBackbone=ResNet502020.07 | 77.2 | 89.4 | 95.6 | — | — | — | |
| MIC2021.02 | 77.2 | 89.4 | 95.6 | — | — | — | |
| Margin-AVSLSetting=512R2022.03 | 76.8 | 89.2 | 95.4 | — | — | — | |
| MSEmbedding size=1282019.04 | 76.6 | 89.2 | 95.2 | 98.4 | — | — | |
| Proxy-AnchorEmbedding size=642020.03 | 76.5 | 89 | 95.1 | 98.2 | — | — |